useful R function

Posted on 2015-05-14

These is a function that I recursively used in analysis using R. I have rewritten in Rcpp to make use of vectorization and make the task faster.

+++

	#include <Rcpp.h>
	using namespace Rcpp;


	// [[Rcpp::export]]
	NumericVector removeNA(NumericVector a)
	{
	int vectorLength = a.size();
	NumericVector result(vectorLength);
	for (int i = 0; i < vectorLength ; i++)
	{
	if (NumericVector::is_na(a[i]))
	{
	result[i] = 0;
	}
	else
	{
	result[i] = a[i];
	}
	}
	return result;
	}

view raw Rfunctions.cpp hosted with ❤ by GitHub

+++

Benchmarking

The function takes in a numeric vector and make all the NA values into 0, which is handy when there's missing data in the table.

I will test the speed compare to the original ifelse R equivalent command.

library(Rcpp)
library(microbenchmark)

#load/compile the functions
sourceCpp('~/scripts/R/Rcpp/Rfunctions.cpp')

#simulate data
a <- sample(c(rnorm(100000,1:100),rep(NA,10000)))

#test second function
microbenchmark(removeNA(a),ifelse(is.na(a),0,a))

## Unit: microseconds
##                    expr       min         lq      mean    median       uq
##             removeNA(a)   665.065   771.2185  2289.387  1122.832  1600.63
##  ifelse(is.na(a), 0, a) 36003.881 43070.1215 62004.357 49713.592 90096.88
##        max neval cld
##   53051.31   100  a 
##  132315.03   100   b

From the experiment, Rcpp functions make use of vectorization and increase the speed significantly (did not do a t-test, but probably p<0.05).